Plant Ecol (2009) 201:517–529 DOI 10.1007/s11258-008-9511-1

Extent and spatial patterns of grass bald land cover change (1948–2000), Oregon Coast Range, USA

Harold S. J. Zald

Received: 8 February 2008 / Accepted: 15 September 2008 / Published online: 2 October 2008 Ó Springer Science+Business Media B.V. 2008

Abstract Globally, temperate grasslands and mead- specific habitat requirements. Tree encroachment ows have sharply declined in spatial extent. Loss and patterns and increased bald edge densities suggest fragmentation of grasslands and meadows may impact increasing rates of bald loss in the future. The remote biodiversity, carbon storage, energy balance, and sensing nature of this study cannot determine the climate change. In the Pacific Northwest region of fundamental causes of bald decline, although prior North America, grasslands and meadows have research suggests climate change, cessation of native declined in extent over the past century. Largely burning, successional changes in response to prior undocumented in this regional decline are the grass wildfires, and cessation of livestock grazing all may balds of the Oregon Coast Range, isolated grasslands in have potential influence. a landscape dominated by coniferous forests. This study was conducted to quantify the spatial extent and Keywords Grassland decline Land cover patterns of grass bald change. Five balds in the Oregon change Forest encroachment Coast Range were evaluated using historical aerial Habitat loss Fragmentation photographs and recent digital orthophoto quadrangles (DOQ). Over the time period of study (1948/1953 to 1994/2000), bald area declined by 66%, primarily from forest encroachment. The number and average size of Introduction bald vegetation patches declined, while edge density increased. Tree encroachment into balds was inversely Declines in the extent of temperate grasslands, mon- related to distance from nearest potential parent trees. tane meadows, and subalpine meadows appear to be a Spatial patterns of bald loss may result from a forest to global ecological phenomenon (Mast et al. 1997; bald gradient of unfavorable environmental conditions Bowman et al. 2001; Didier 2001; League and Veblen for tree establishment and/or seed dispersal limitation. 2006; Marie-Pierre et al. 2006; Baker and Moseley Species dependent on balds may be at risk from loss of 2007; Coop and Givnish 2007). Noting that conver- bald area and increased fragmentation, although met- sions of temperate grasslands to other land cover types rics of habitat fragmentation may not reflect species- greatly exceeds conservation; Hoesktra et al. (2005) consider the loss of grasslands as a major component of a global ‘‘biome crisis’’. Extensive losses of & H. S. J. Zald ( ) grassland habitat will reduce biodiversity, since hab- Forest Science Department, Oregon State University, 321 Richardson Hall, Corvallis, OR 97331, USA itat modification and fragmentation are believed to be e-mail: [email protected] significant threats to biodiversity (Sala et al. 2000; 123 518 Plant Ecol (2009) 201:517–529

Foley et al. 2005). Grasslands and meadows occupy a of bald change may provide information to make significant portion of global land surface, and are inferences about the underlying proximate processes major contributors to biosphere primary production involved. Landscape metrics of fragmentation are a (Field et al. 1998; Saugier et al. 2001). Large-scale widely used proxy for habitat degradation. Mapping declines of grasslands and meadows may change of bald change, and the extent of bald fragmentation regional carbon storage (Houghton et al. 1999; Jack- could provide land managers with useful spatially son et al. 2002; Knapp et al. 2008), and surface albedo explicit information for establishing reference condi- (Bonan et al. 1992), thereby influencing global cli- tions and prioritizing specific areas for restoration mate change. activities. In the Pacific Northwest Region of North America, The objectives of this study were to: (1) quantify the the loss of lowland, montane, and subalpine grasslands extent of balds in the Oregon Coast Range over the past have been largely attributed to tree encroachment five decades (1948–2000), (2) determine what types of (Habeck 1961; Franklin et al. 1971; Woodward et al. land cover balds are being converted to (i.e., coniferous 1995; Rochefort and Peterson 1996; Miller and Halp- forest, roads and building, bare ground, etc.), (3) ern 1998; Foster and Shaff 2003), although agricultural quantify the spatial patterns of tree encroachment into development has played a major role in the loss of the balds, and (4) determine if bald vegetation patches Willamette Valley prairies (Habeck 1961; Johannessen have become more fragmented and edgier (i.e., decline et al. 1971). Largely overlooked within the region have in total number of patches and average patch size, been the grass balds of the Oregon Coast Range, while having increased edge densities) over the time graminoid dominated vegetation which typically occur period of study. at the summits of Coast range peaks and contrast strikingly in structure with the taller surrounding coniferous forests (Franklin and Dyrness 1998). Like Methods balds of the Southern Appalachian Mountains of North America, these balds had been relevantly stable in Study area extent, while containing rare and endemic plant species, and disjunct populations, suggesting they are Franklin and Dyrness (1998) listed seven major balds in relict vegetation assemblages (Merkle 1951; Detling the Oregon Coast Range, but omitted the extensive bald 1953; Wiegl and Knowles 1995). However, there is on Mount Hebo (USFWS 2001). Of these eight balds, considerable uncertainty regarding the origins, main- only five were used in this study due to the poor quality tenance, and decline of the balds, or the role of humans of historical imagery at three balds (Fig. 1). From north in bald dynamics. Coast Range balds provide habitat to tosouth, balds studied are: Mount Hebo, BaldMountain, multiple species with varying levels of Federal and Marys Peak, Grass Mountain, and Praire Peak (see State protected status (USFWS 2001; ONHIC 2007). Table 1 for study area information). Bald elevations Their limited spatial extent, unique vegetation, and range from 968 to 1249 m. All five balds are in the presence of disjunct and rare species contribute to their Tsuga heterophylla zone, Coast Range Level III Eco- ecological importance and value to society. region (Franklin and Dyrness 1998; Thorson et al. Limited research within the balds has focused on 2003). However, higher balds (such as Marys Peak) may vegetation classification (Merkle 1951; Aldrich 1972; lie within the higher elevation, Abies amabilis zone Snow 1984). Magee and Antos (1992) investigated (McCain and Diaz 2002). The climate is maritime, with tree encroachment into one Coast Range bald, but no mild wet winters and cool dry summers, but climate published documentation exists quantifying current varies with proximity to the ocean, latitude, and and historical bald extent, or changes in bald area over orographic effects. Over the 1970–2000 time period, time. There is no documentation of the types of bald climatic data from the nearest weather station (Laurel loss (i.e., forest encroachment versus human distur- Mountain Weather Station, National Weather Service bance), spatial patterns of bald loss, or landscape Cooperative Network #354776, 44°550 N123°340 W, metrics (i.e., number of bald patches, area of patches, 1090 m), reported annual average maximum tempera- and edge density of patches) of bald vegetation ture of 10°C, mean temperature of 6.6°C, annual fragmentation over time. Quantifying spatial patterns average minimum temperatures of 3.1°C, annual 123 Plant Ecol (2009) 201:517–529 519

Bald vegetation varied between study areas. Common graminoid species include: Elymus glaucus, Festuca rubra, Bromus carinatus, Poa pratensis (non-native), Agrostis diegoensis, Carex californica,andCarex rossii. Common forbs and ferns include: Lomatium martendalei, Collinsia parviflora, Senecio triangularis, Delphinium menziesii, Iris tenax, and Pteridium aquilinum (see, Merkle 1951; Aldrich 1972; Schuller and Exeter 2007 for details). Disjunct populations of more typi- cally Eastern Cascades species include Festuca idahoensis and Carex hoodii, while populations of species more typical of high altitudes include Phleum alpinum, Arctostaphylos uva-ursi, Phlox cepitosa, Silene douglasii, Allium crenulatum, and Lupinus lepidus var. lobbi (Merkle 1951; Detling 1953; Oregon Plant Atlas 2007). Multiple species of concern are associated with grass bald vegetation in the Oregon Coast Range, including: the Federally listed Oregon Silverspot Butterfly (Speyeria zerene hippolyta), the State listed Coast Range Fawn Lilly (Erythronium elegans), and three plants that are candidates for State protection (Cardamine pattersonii, Saxifraga hitch- cockiana, and Sidalcea hirtipes) (USFWS 2001; ONHIC 2007). Nomenclature of plant species follows Hitchcock and Cronquist (1976). Multiple conifer species can encroach into the balds. At Mount Hebo Psuedotsuga menziesii dominates encroachment (personal observation). P. menziesii is also the dom- inant encroaching species at Bald Mountain and Prairie Peak. P. menziesii, Abies procera, and to a lesser degree Tsuga heterophylla are encroaching at Grass Mountain, while A. procera is the dominant encroach- ing species at Marys Peak (Magee and Antos 1992; Schuller and Exeter 2007). Fig. 1 Documented major grass balds within the Oregon Coast Range ecoregion. Grass Balds names are: SM—Saddle Aerial photo change detection Mountain, MH—Mount Hebo, BM—Bald Mountain, MP— Marys Peak, GM—Grass Mountain, PP—Prairie Peak, RN— Roman Nose Mountain, and TY—Tyee Mountain. Grass Balds Analyses used historical (1948 and 1953) aerial with solid triangles were studied photographs and recent (years 1994 and 2000) digital orthophoto quadrangles (DOQ’s) of the five study areas. The historical images were geo-rectified using precipitation of 307 cm, and total snowfall of 301.5 cm the DOQ’s as reference imagery. Historical photos and which falls predominantly from November to April. DOQ’s were filtered to improve classification of Mount Hebo, Bald Mountain, and Prairie Peak study imagery into distinct land cover classes. Filtered areas are predominantly underlain by mafic intrusions of images were manually classified into four land cover Oligocene origin, while Marys Peak and Grass Moun- classes (bald, coniferous forest, roads and buildings, tain study areas are underlain by a combination of and bare ground) within each image. A matrix function Oligocene mafic intrusions and middle Eocene sand- was applied to these classified images, resulting in a stones and siltstones (Walker and MacLeod 1991). single image with 16 (4 9 4) potential land cover 123 520 Plant Ecol (2009) 201:517–529

Table 1 Grass bald study Study area Elevation (m) Latitude Longitude Imagery dates (dd/mm/year) area maximum elevations, geographic coordinates, and Mount Hebo 968 45°120 N 123°450 W 29/07/1953, 30/05/1994 imagery dates Bald Mountain 985 44°470 N 123°320 W 29/06/1948, 08/08/2000 0 0 a Two historical images Marys Peak 1,249 44°30 N 123°33 W 14/07/1948, 30/08/1994 were mosaicked for Prairie Grass Mountain 1,098 44°270 N 123°330 W 29/06/1948, 30/05/1994 Peak. See Appendix A for Prairie Peak 1,043 44°160 N 123°370 W 29/06/1948a, 02/07/1948a, 30/08/1994 details change classes for each study area. Many of these 16 landscape area (m2) of all bald patches, multiplied by change classes were irrelevant for this study (such as 10,000 (to convert to meters of edge per hectare). coniferous forests converted to roads and buildings), so Higher edge density values are indicative of complex the numbers of land cover classifications were reduced or elongated patches, lower edge density values to aid in visual interpretation of bald change (Fig. 2). indicate more compact or simple shaped patches. The surface area of the remaining relevant cover The function Distance in the Spatial Analysis of classes were compiled to determine the amount of ArcGIS 9.1 (2005 ERSI Inc.) was used to test the historical and current bald area, the extent of bald relationship between tree encroachment and distance to change, and what types of land cover balds were being potential parent trees. The Euclidean distance was converted to (i.e., coniferous forest, roads and build- calculated between each pixel of bald encroached by ings, and bare ground). Unless otherwise mentioned, trees and the nearest pixel of potential parent tree (i.e., image processing and analyses took place using nearest forested pixel in the historical imagery). ERDAS Imagine 8.7 (2003 Leica Geosystems GIS Euclidean distance was also calculated between each and Mapping, LLC). For a detailed description of pixel of bald that remained stable (did not change image acquisition and processing methods please refer between imagery dates) and the nearest pixel of to Appendix A. potential parent tree. The percentage of pixels within both encroached bald and unchanged bald classes were separately rescaled to total 100%, and plotted with Spatial pattern metrics respect to distance from nearest potential parent trees. A Wilcoxon signed rank sum test was used to determine Landscape metrics were calculated to quantify bald whether distribution of pixels with respect to distance patch fragmentation. The number of bald patches, from potential parent trees differed between encroached average patch size (referred to as patch size), area- bald and unchanged bald pixels. Distributions were weighted average patch size (referred to as weighted considered different if the Wilcoxon signed rank sum patch size), and edge density were calculated for both statistic (S value) was significant at a = 0.05 level. historical and recent classified images. Weighted patch size was calculated in addition to simple patch size because the distribution of bald patch sizes was skewed to a few large patches and many small patches. In this Results patch distribution, the simple arithmetic average will not reflect the expected patch size encountered by Extent and decline of grass balds random placement of points, resulting in a dispropor- tionate effect of small patches (Turner et al. 2001). The All five balds declined in area over the time period of weighted patch size was calculated as, study (Table 2). From 1948/1953 to 1994/2000, balds X .X declined from 523.6 ha to 179.2 ha (66%). Forest 2 encroachment into balds was the dominant type of Sa ¼ Sk ðÞSk ; bald decline (348.7 ha, 95%). Conversion of balds to where Sa is the area-weighted average patch size, and roads and buildings accounted for 15.4 ha (4%), Sk is the size of the kth patch. Edge density was while conversion to bare ground was minor. calculated as the sum of the lengths (m) of all edge Mount Hebo displayed the greatest bald decline. segments within the bald patches divided by the Bald Mountain was the smallest historical bald, had 123 Plant Ecol (2009) 201:517–529 521

Fig. 2 Example (Mount Hebo study area) of aerial photo grass bald change detection process. a 1953 historical aerial photo, b 1994 DOQ, c 1953 classified image, d 1994 classified image, and e matrix image depicting grass bald change at Mount Hebo from 1953 to 1994

the lowest proportional bald decline, and the greatest converting the contiguous main bald into a series of proportional amount of bald area decline due to road smaller interconnected balds (not shown). and building development. Visual interpretation of the Bald Mountain imagery (not shown) suggests the Spatial patterns of grass bald change 0.2 ha conversion of bald to bare ground resulted from adjacent logging operations in the southern edge Landscape metrics of bald fragmentation (number of of the study area. Prairie Peak had the second highest bald patches, patch size, weighted patch size, and percentage of bald decline. Bald decline at Prairie edge density) were generally consistent between Peak resulted in the near complete elimination of one study areas (Table 3). From 1948/1953 to 1994/ bald in the eastern part of the study area, while 2000, the number of bald patches declined at all study

123 522 Plant Ecol (2009) 201:517–529

Table 2 Grass bald changes by area and types of land cover change Study area Date Area (ha) Type changea D (ha)b D (%)c

Mount Hebo 1953 243.4 Bald to forest 202.1 96.09 1994 39.8 Bald to bare ground 0.2 0.10 1953–1994 -203.6 Bald to R&B 8.0 3.79 Conversion to bald 6.6 Bald Mountain 1948 24.3 Bald to forest 14.8 84.8 2000 7.3 Bald to bare ground 0.2 1.2 1948–2000 -17.1 Bald to R&B 2.4 14.0 Conversion to bald 0.4 Marys Peak 1948 118.4 Bald to forest 47.0 93.9 1994 77.2 Bald to bare ground 0.4 0.7 1948–1994 -41.1 Bald to R&B 2.7 5.5 Conversion to bald 9.0 Grass Mountain 1948 43.9 Bald to forest 22.0 98.3 1994 23.1 Bald to bare ground 0.0 0.0 1948–1994 -20.8 Bald to R&B 0.4 1.9 Conversion to bald 1.6 Prairie Peak 1948 93.6 Bald to forest 62.7 95.4 1994 31.8 Bald to bare ground 1.1 1.7 1948–1994 -61.8 Bald to R&B 1.9 2.8 Conversion to bald 3.9 All Study Areas 1948/1953 523.6 Bald to forest 348.7 95.3 1994/2000 179.2 Bald to bare ground 1.9 0.5 Change -344.4 Bald to R&B 15.4 4.2 Conversion to bald 21.6 – a Type change refers to what type of land cover balds are converted to. R&B refers to roads and buildings b D (ha) refers to the area (in hectares) of each type of land cover change c D (%) refers to the proportion of bald loss due to a specific type of land cover change Note: Bald area values used to calculate D(ha) and D(%) are bald loss values corrected for conversions of non-bald land cover classes to bald vegetation areas (47% total 10–67% for individual study areas). -2184, P \ 0.0001), Marys Peak (S =-8333.5, Patch size for four out of the five balds declined (49% P \ 0.0001), Grass Mountain (S =-1557, average, 20–66% for individual study areas). The P \ 0.0001), and Prairie Peak (S =-908.5, fifth study area (Marys Peak) saw a 2% increase in P \ 0.0001). Results for Bald Mountain (S = patch size. Weighted patch size declined sharply in -876.5, P = 0.0836) suggest a similar relationship, all study areas (94% average, 48–97% for individual but also displayed longer distance tree encroachment. study areas). Bald edge density increased at four out of the five study areas (18% total, 7–46% for individual study areas). The fifth study area (Marys Discussion Peak) saw only a 0.41% increase in edge density. Forest encroachment into the balds was inversely Grass bald decline and fragmentation related to the distance from potential parent trees (Fig. 3). The distribution of encroached bald versus This study is the first to document the widespread unchanged bald in relation to distance from potential decline and fragmentation of grass balds in the parent trees was significant at Mount Hebo (S = Oregon Coast Range. All five study areas displayed 123 Plant Ecol (2009) 201:517–529 523

Table 3 Landscape Study area Date No. of bald Patch size Weighted patch Edge density metrics of grass balds by patches (m2) (m2) size (m2) (m2/ha) study area and date Bald Mountain 1948 87 2,945 226,910 15,021 2000 78 1,352 24,893 19,354 Grass Mountain 1948 778 595 66,473 14,957 1994 515 473 34,398 15,983 Marys Peak 1948 1,651 791 597,727 14,660 1994 1,066 809 202,988 14,719 Mount Hebo 1953 2,446 1,119 2199,737 13,911 1994 1,324 380 63,454 16,081 Prairie Peak 1948 2,025 724 788,861 9,563 1994 674 538 64,133 13,996 See section ‘‘Methods’’ for All study areas 1948/1953 6,987 5,384 3281,981 68,111 detailed descriptions of 1994/2000 3,657 2,744 186,878 80,134 landscape metrics large losses in bald area due to tree encroachment, suggests bald decline was largely due to the elimination with only 35% of 1948/1953 bald area present in of small patches, and the shrinking of larger patches 1994/2000. Study areas displayed declining numbers without becoming very complex or elongated in shape. of bald patches, reduced patch sizes and weighted patch sizes, and increased edge densities. However, Spatial patterns and proximate causes of tree these metrics of fragmentation varied by study area, encroachment suggesting different patterns of bald change. At Bald Mountain there were small declines in the number of Bald decline was overwhelming due to tree patches, large declines of patch size and weighted encroachment, which was greatly closer to the forest patch size, and large increases in edge density. This is edge. This infilling pattern of encroachment is consistent with the elimination of only a few small consistent with field data from Marys Peak (Magee bald patches, but larger declines in the area of larger and Antos 1992), but differs from more clustered patches, and edgier (i.e., more complex and/or patterns of encroachment found in montane and elongated) remnant balds. Mount Hebo and Prairie subalpine meadows of the Oregon and Washington Peak had the greatest declines in the number of bald Cascades (Franklin et al. 1971; Miller and Halpern patches, patch size, and weighted patch size, but had 1998). different changes in edge density. Prairie Peak bald The spatial patterns of tree encroachment likely patches had a much greater increase in edge density result from a combination of unfavorable tree estab- compared to Mount Hebo, suggesting that while both lishment conditions due to high resource competition of these study areas experience declines in both large from graminoids, and dispersal limitation of tree seed and small bald patches, the remnant patches are with increased distance from the forest edge. The Prairie Peak have more elongated or complex shapes influence of forest edge on non-forest microsite than those at Mount Hebo. In contrast, Marys Peak conditions decline with increased distance from the had a large reduction in the number of bald patches, a forest edge (Chen et al. 1993). With increased distance slight increase in patch size, a large reduction in from the forest edge there is increased graminoid weighted patch size, and almost no increase in edge cover. For many conifer species, both establishment density. These changes are consistent with the and growth decline with increased graminoid cover predominant disappearance of smaller patches, and (Alan and Lee 1989; Kolb and Robberecht 1996; reductions of remaining large patches areas while Dovciak et al. 2008). A. procera germinants and maintaining less complex and/or elongated shapes. seedlings tend to be at a competitive disadvantage Grass Mountain had a small increase in edge density, against dense bald vegetation (Magee and Antos 1992). and when combined with the other patch metrics, Pseudotsuga menziesii and Tsuga heterophylla

123 524 Plant Ecol (2009) 201:517–529

Fig. 3 Distribution of encroached versus unchanged grass bald land cover in relation to distance from potential parent trees

establishment and growth also decline with increases driven by a multivariant gradient from the forest to in herbaceous vegetation competition (Rose et al. core grassland of seed dispersal and environmental 1999; Rose and Ketchum 2002). Wind-dispersed seeds conditions (overstory cover, moss and grass cover, can travel quiet far (Clark et al. 1999), but seed etc.). Although specific microsite information is dispersal displays inverse seed density with increased lacking for this study, prior studies within the grass distance from source trees. The heavy weight of balds and in other grasslands suggest both seed A. procera seeds results in a relatively short dispersal dispersal and environmental conditions appear to distances compared to P. menziesii and T. heterophylla. drive spatial patterns of forest encroachment. However, almost all bald vegetation was within 100 m of pre-existing trees, well within dispersal distances of Ecological implications invading species (Isaac 1930; Franklin and Smith 1974; Carkin et al 1978). Multiple species of concern are associated with bald In the Carpathian Mountains, Dovciak et al. (2008) vegetation in the Oregon Coast Range (USFWS 2001; found invasion of grasslands by Picea abies to be ONHIC 2007). Declines in habitat area and increased 123 Plant Ecol (2009) 201:517–529 525 fragmentation are generally associated with the loss of study cannot provide empirical evidence regarding its species diversity and reductions in species populations fundamental causes. However, a review of regional (Saunders et al 1991; Sala et al. 2000; Krauss et al. causes of tree encroachment, combined with site 2004; Foley et al. 2005). Although, landscape metrics, specific disturbance and land use history, may such as habitat area and fragmentation can be impor- provide valuable information. Regional causes of tant, they cannot directly quantify habitat quality or tree encroachment vary by study, but most often species composition. Reliance on landscape metrics is include: climate change, fire and fire suppression, often necessary in the absence of, but may be less grazing history, and succession. important than, species-specific field-based measures Climate changes (via changes in growing season of habitat suitability (Hobbs and Yates 2003; Fischer length, snow free periods, growing season tempera- and Lindenmayer 2007). Additionally, reductions in tures, etc.) are a likely cause of regional subalpine species populations and species richness may lag tree encroachment (Franklin et al. 1971; Woodward temporally behind habitat loss, further complicating et al. 1995; Rochefort and Peterson 1996). Tree the relationship between species and habitat (Tillman establishment in the balds of Marys Peak coincided et al. 2002; Helm et al. 2006). Despite these uncer- with favorable climatic conditions (Magee and Antos tainties, large declines in bald area and increased bald 1992). However, has previously noted, Marys Peak is fragmentation are likely to negatively impact rare at a higher elevation and contains a more subalpine species and disjunct populations. forest type than other studied balds, so conifer Patterns of tree encroachment and bald fragmen- establishment–climate relationships at Marys Peak tation suggest a positive feedback of accelerating may not be representative of other lower elevation encroachment in the future. In some meadow types, balds. Livestock grazing also can be an important tree establishment positively alters microsite condi- driver of tree encroachment, reducing plant cover and tions with respect to future tree regeneration and density, while exposing bare soil. Miller and Halpern establishment (Miller and Halpern 1998; Dovcˇiak (1998) concluded that tree encroachment in some et al. 2008). Since the remnant bald patches are closer upland meadows in the Oregon Cascades occurred to forest edge, they are more likely to encounter a following grazing cessation and coincidently, favor- higher density of tree seed. Higher edge density of able climatic conditions. Homesteaders grazed remnant bald patches also suggests the surrounding livestock on the balds from the 1800s to 1940s forest matrix will have increased influence on eco- (Reynolds 1993), so observed forest encroachment logical patterns and processes in the remnant patches could be partly due to grazing cessation prior to (Saunders et al. 1991). In particular, increased shade historical imagery from the forest edge may reduce graminoid cover that High-severity wildfires are an important distur- excludes conifer establishment. If conservation and bance agent in Coast Range forests that has varied restoration of balds is desired, the potential for considerably over time. About 4600 to 2700 years additional and accelerating bald loss should be a BP, the Oregon Coast Range used to have a management concern. Bald decline due to road and 140 ± 30 year fire intervals, compared to building development was minor at most study areas 240 ± 30 year from 2700 years ago to present (Long (except the Bald Mountain study area). However, and Whitlock 2002). Native Americans are believed landscape metrics used in this study cannot quantify to have used frequent low-severity fire to manage changes in habitat quality that occur even if the land some of the Coast Range grass balds prior to Euro- cover type remains unchanged, while roads can American homesteading (personal communication increase the dispersal and establishment of non-native Phyllis Steeves, Archeologist, Siuslaw National For- plant species (Parendes and Jones 2000; Trombulak est). By 1860, disease-driven declines in Native and Frissell 2000; Christen and Matlack 2006). populations likely resulted in dramatic reductions of Native fire ignitions on the grass balds. Elimination Fundamental causes of bald decline of Native American ignitions, combined with active fire suppression, has driven tree encroachment in the The proximate cause of bald decline is tree encroach- prairies of the Puget Sound and Willamette Valley ment, but the primary objectives and data of this (Johannessen et al. 1971; Foster and Shaff 2003; 123 526 Plant Ecol (2009) 201:517–529

Boyd 1999). Evidence of extensive fires in the region Map Library in Eugene, OR. All historical aerial could suggest grass balds are early-successional photographs were taken during the summer of 1948 vegetation that has since been invaded after over (except Mount Hebo, taken during the summer of 50 years of fire suppression. However, there is strong 1953). Historical aerial photographs were scanned on evidence balds have been long term features on the a flatbed scanner at 600 dpi. Scanned historical landscape: balds generally lack biological legacies images had 8-bit radiometric resolution and 0.63– (i.e., large woody debris, stumps, snags, and remnant 0.89 m postrectification spatial resolution, re-sam- trees) present even after multiple disturbances, they pled to 1 m pixel size after geo-rectification using are dominated by graminoids rather than shrub- nearest neighbor interpolation. Recent images of the dominated early-successional vegetation typical in study areas are panchromatic digital orthophoto the Coast Range, and contain disjunct and endemic quadrangles (DOQs). DOQs were downloaded from plant species suggestive of long term vegetation TerraServer-USA (http://terraserver-usa.com/default. stability (Merkle 1951; Detling 1953; Dyrness 1973; aspx). The Mount Hebo, Marys Peak, Grass Moun- Hemstrom and Logan 1986; Franklin et al. 2002). tain, and Prairie Peak DOQs were taken in the One potential exception is Mount Hebo, where high summer of 1994 (except Bald Mountain, taken during intensity fires, active reforestation, commissioning the summer of 2000). DOQs were in Universal and decommission of the Mount Hebo Air Station, Transverse Mercator Projection (UTM, Zone 10N), along with mowing and burning to maintain habitat using the Geodetic Reference System Spheroid of for the Oregon Silverspot Butterfly, create a very 1980, and the North American Datum of 1983. All complex disturbance history (Munger 1944; Oregon DOQs had 8-bit radiometric resolution and 1 m Division of Forestry 2001; USFWS 2001). Climate spatial resolution. change and complex disturbance history likely inter- act to shape spatial and temporal patterns of tree invasion into the grass balds, as found elsewhere Aerial photo geo-rectification (Miller and Halpern 1998). However, addressing these fundament drivers of tree invasion is beyond the Each scanned historical image was first manually scope of this study. Spatially-explicit tree age and clipped so only grass balds and surrounding forest growth chronology data, combined with climate vegetation were retained. Each clipped image was records and detailed grazing history, may allow geo-rectified using the recent DOQ of the same site inference regarding the fundamental causes of tree as the reference image. Control Points (CPs) between invasion into these unique landscape features. historic and recent images were manually selected, focusing on the most temporally stable features (i.e., Acknowledgments The author acknowledges Colin Kelly at buildings, roads, rock outcrops, and individual open the University of Oregon Map Library for assistance locating grown trees). A minimum of 20 CPs between and organizing hardcopy historical aerial photographs. Special thanks to Jonathan Thompson for helpful advice regarding historical and reference imagery was desired, image enhancement and spatial statistics. Funding for this study although this was not possible in all cases. Geo- was provide by the U.S. Forest Service, Pacific Northwest rectification models used polynomial equations, Forest Inventory and Analysis Program. Additional thanks to evaluated by the Root Mean Squared Error (RMSE), Tom Spies, Andy Gray, Rob Pabst, and two anonymous reviewers for providing thoughtful comments and suggestions and rejected if the RMSE exceeded 0.5 pixels. on prior versions of this manuscript. Additionally, rectified images were digitally overlain as partial transparencies over recent DOQs to qualitatively assess rectification accuracy. Rectifica- Appendix A. Aerial photograph acquisition tion and resampling occurred using the nearest and image processing neighbor method, conserving raw image brightness values. A mosaic of two rectified images was Aerial photo acquisition required for the Prairie Peak study area (Prairie Peak a & b), and was generated using histogram Historical panchromatic 1:12,000 scale aerial photo- matching, no cutlines, minimum select function, and graphs were acquired from the University of Oregon union of all inputs. 123 Plant Ecol (2009) 201:517–529 527

Study area Date CPs RMSE X RMSE Y RMSE total Polynomial order

Mount Hebo 1953 22 0.05 0.03 0.06 1st Bald Mountain 1948 8 0.03 0.01 0.02 1st Marys Peak 1948 11 0.02 0.29 0.04 2nd Grass Mountain 1948 20 0.02 0.01 0.02 2nd Prairie Peak—a 1948 24 0.06 0.06 0.08 1st Prairie Peak—b 1948 20 0.04 0.02 0.05 1st

Image filtering, level slicing land cover where small areas of forest were misclassified as grass classification, and matrixing balds. Clumping and sieving was used to reclassify these false balds as forest. An eight pixel window was DOQs and rectified historical images were filtered used to clump contiguous groups of bald pixels into using a 5 9 5 pixel focal minimum filter followed by a individual bald patches. Bald patches were determined 5 9 5 low-pass filter. This filtering combination com- manually by overlaying the clumped bald patches over pensated for highly reflective (high brightness value) the associated rectified image or DOQ. Clumped bald pixels within continuous forest canopy that would be patches that were smaller than the smallest visually misclassified during land cover classification as bald determined patches within a historical or DOQ image vegetation or bare ground unless their brightness values were reclassified as forest, resulting in historical and were reduced. The 5 9 5 focal minimum filter recent 4-class (forest, meadow, bare ground, and roads assigned the lowest brightness value of the neighboring and buildings) land cover images for each study area. 24 pixels to the central pixel. Focal filtering was Change detection of bald land cover 4 between followed by a 5 9 5 low-pass filter to deemphasize historical and recent classified images used a matrix high-frequency edges (Lillesand et al. 2004). function, creating a new image showing the coinci- Classification of pixels into three landcover classes dence of values between the historical and recent (grass bald, forest, and bare ground) was accomplished classified images. Each matrix classified image had 16 by level slicing of gray-level brightness values. This potential coincident classes (4 classes 9 4 classes). method has previously been used on panchromatic These classes were recoded, and the surface area of images to classify pine forests and grasslands in the each land cover type was calculated from the number Colorado Front Range (Mast et al. 1997). Each filtered of pixels within each class. image was sliced individually, and brightness value ranges for each land cover class varied between each of the 10 images. Attempts to classify conifers and hardwoods separately were unsuccessful; instead a References single forest land cover classification included both conifers and hardwoods. Using level slicing to classify Alan RB, Lee WG (1989) Seedling establishment microsites of the roads and buildings land cover class greatly exotic conifers in Chionochloa rigida tussock grassland, reduced classification accuracy of the other three Otago, New Zealand. N Z J Bot 27:491–498 Aldrich FT (1972) A chronological analysis of the grass balds classes. Instead, the roads and buildings were manually in the Oregon Coast Range. Ph.D Dissertation, Oregon delineated as polygons in separate shapefiles for each State University current and rectified historical image in ArcGIS 9.1 Baker BB, Moseley RK (2007) Advancing treeline and (2005 ERSI Inc.). Polygon shapefiles were converted retreating glaciers: implications for conservation in Yun- nan, P. R. China. Arct Antarct Alp Res 39:200–209 to raster images and incorporated into each three-class Bonan GB, Pollard D, Thompson SL (1992) Effects of boreal classified image using an overlay function, resulting in forest vegetation on global climate. Nature 359:716–718 four-class classified images. Bowman DMJS, Walsh A, Milne DJ (2001) Forest expansion Eucalyptus Despite focal minimum and low-pass filtering, and grassland contraction within a savanna matrix between 1941 and 1994 at Litchfield National Park visual comparison of classified imagery to the geo- in the Australian monsoon tropics. Glob Ecol Biogeogr rectified images and DOQs found numerous instances 10:535–548 123 528 Plant Ecol (2009) 201:517–529

Boyd R (1999) Strategies of Indian burning in the Willamette structural development of natural forest ecosystems with Valley. In: Boyd R (ed) Indians fire and the land in the silvicultural implications, using Douglas-fir forests as an Pacific Northwest. Oregon State University Press, Cor- example. For Ecol Manag 155:399–423 vallis, OR, pp 94–138 Habeck JR (1961) The original vegetation of the Mid- Carkin RE, Franklin JF, Booth J, Smith CE (1978) Seeding Willamette Valley, Oregon. Northwest Sci 35:65–77 habits of upper-slope tree species IV: seed flight of noble Helm A, Hanski I, Pa¨rtel M (2006) Slow response of plant fir and Pacific silver fir. USDA Forest Service Research species richness to habitat loss and fragmentation. Ecol Notes PNW-312. Pacific Northwest Forest and Range Lett 9:72–77 Experimental Station, Portland, OR Hemstrom MA, Logan SE (1986) Plant association and man- Chen J, Franklin JF, Spies TA (1993) Contrasting microcli- agement guide Siuslaw National Forest. USDA Forest mates among clear-cut, edge, and interior of old-growth Service Pacific Northwest Region R6-Ecol 220-1986a. Douglas-fir forest. Agric For Meteorol 63:219–237 Available via http://www.reo.gov/ecoshare/Publications/ Christen D, Matlack G (2006) The role of roadsides in plant searchresults.asp. Accessed 8 Feb 2008 invasions: a demographic approach. Conserv Biol 20:385– Hitchcock CL, Cronquist A (1976) Flora of the Pacific 391 Northwest. University of Washington Press, Seattle Clark JS, Silman M, Kern R, Macklin E, HilleRisLambers J Hobbs RJ, Yates CJ (2003) Impacts of ecosystem fragmenta- (1999) Seed dispersal near and far: patterns across tem- tion on plant populations: generalizing the idiosyncratic. perate and tropical forests. Ecology 80:1475–1494 Aust J Bot 51:471–488 Coop JD, Givnish TJ (2007) Spatial and temporal patterns of Hoesktra JM, Boucher TM, Ricketts TH, Roberts C (2005) recent forest encroachment in montane grasslands of the Confronting a biome crisis: global disparities of habitat Valles Caldera, New Mexico, USA. J Biogeogr 34:914–927 loss and protection. Ecol Lett 8:23–29 Detling LE (1953) Relict islands of xeric flora west of the Houghton RA, Hackler JL, Lawrence KT (1999) The U.S. Cascade Mountains in Oregon. Madrono 12:39–47 carbon budget: contributions from land-use change. Sci- Didier L (2001) Invasion patterns of European larch and Swiss ence 285:574–578 stone pine in subalpine pastures in the French Alps. For Isaac LA (1930) Seed flight in the Douglas-fir region. J For Ecol Manag 145:67–77 28:492–499 Dovcˇiak M, Hrivna´k R, Ujha´zy K, Go¨mo¨ry D (2008) Seed rain Jackson RB, Banner JL, Jobba´gy EG, Pockman WT, Wall DH and environmental controls on invasion of Picea abies (2002) Ecosystem carbon loss with woody plant invasion into grassland. Plant Ecol 194:135–148 of grasslands. Nature 418:623–626 Dyrness CT (1973) Early stages of plant succession following Johannessen CL, Davenport WA, Millet A, McWilliams S logging and burning in the Western Cascades of Oregon. (1971) The vegetation of the Willamette valley. Ann Ecology 54:57–69 Assoc Am Geogr 61:286–302 Field CB, Behrenfeld MJ, Randerson JT, Falkowski P (1998) Knapp AK, Briggs JM, Collins SL, Archer SR, Bret-Harte MS, Primary production of the biosphere: integrating terrestrial Ewers BE, Peters DP, Young DR, Shaver GR, Pendall E, and oceanic components. Science 281:237–240 Cleary MB (2008) Shrub encroachment in North Ameri- Fischer J, Lindenmayer DB (2007) Landscape modification and can grasslands: shifts in growth from dominance rapidly habitat fragmentation: a synthesis. Glob Ecol Biogeogr alters control of ecosystem carbon inputs. Glob Change 16:265–280 Biol 14:615–623 Foley JA, DeFries R, Asner GP, Barford C, Bonan G, Car- Kolb PF, Robberecht R (1996) Pinus ponderosa seedling estab- penter SR, Chapin FS, Coe MT, Daily GC, Gibbs HK, lishment and the influence of competition with the Helkowski JH, Holloway T, Howard EA, Kucharik CJ, bunchgrass Agropyron spicatum. Int J Veg Sci 157:509–515 Monfreda C, Patz JA, Prentice IC, Ramankutty N, Snyder Krauss J, Klein A-M, Steffan-Dewenter I, Tscharntke T (2004) PK (2005) Global consequences of land use. Science Effects of habitat area, isolation, and landscape diversity 309:570–574 on plant species richness of calcareous grasslands. Bio- Foster JR, Shaff SE (2003) Forest colonization of Puget divers Conserv 13:1427–1439 Lowland grasslands at Fort Lewis, Washington. North- League K, Veblen T (2006) Climatic variability and episodic west Sci 77:283–296 Pinus ponderosa establishment along the forest-grassland Franklin JF, Dyrness CT (1998) Natural vegetation of Oregon ecotones of Colorado. For Ecol Manag 228:98–107 and Washington. Oregon State University Press, Corval- Lillesand TM, Kiefer RW, Chipman JW (2004) Remote lis, OR sensing and image interpretation, 5th ed. John Wiley and Franklin JF, Smith CE (1974) Seeding habits of upper-slope Sons, Inc., New York, 763 pp tree species II: dispersal of a mountain hemlock seedcrop Long CJ, Whitlock C (2002) Fire and vegetation history from on a clearcut. USDA Forest Service Research Note PNW- the coastal rain forest of the Western Oregon Coast 214. Pacific Northwest Forest and Range Experimental Range. Quatern Res 58:215–225 Station, Portland, OR Magee TK, Antos JA (1992) Tree invasion into a mountain-top Franklin JF, Moir WH, Douglas GW, Wiberg C (1971) Inva- meadow in the Oregon Coast Range, USA. J Veg Sci sion of subalpine meadows by trees in the Cascade Range, 3:485–494 Washington and Oregon. J Arct Alp Res 3:215–224 Marie-Pierre J, Didier A, Gerard B (2006) Patterns of ash Franklin JF, Spies TA, Van Pelt R, Carey AB, Thornburgh DA, (Fraxinus excelsior L.) colonization in mountain grass- Berg DR, Lindenmayer DB, Harmon ME, Keeton WS, lands: the importance of management practices. Plant Shaw DC, Bible K, Chen J (2002) Disturbances and Ecol 183:177–189 123 Plant Ecol (2009) 201:517–529 529

Mast JN, Veblen TT, Hodgson ME (1997) Tree invasion within Sala OE, Chapins FSIII, Armesto JJ, Berlow E, Bloomfield J, a pine/grassland ecotone: an approach with historical Dirzo R, Huber-Sanwald E, Huenneke LF, Jackson RB, aerial photography and GIS modeling. For Ecol Manag Kinzig A, Leemans R, Lodge DM, Mooney HA, Oester- 93:181–194 held M, Poff NL, Sykes MT, Walker BH, Walker M, Wall McCain C, Diaz N (2002) Field guide to the forested plant DH (2000) Global biodiversity scenarios for the year associations of the Northern Oregon Coast Range. USDA 2100. Science 287:1770–1774 Forest Service, Pacific Northwest Region. Technical Saugier B, Roy J, Mooney HA (2001) Estimations of global Paper R6-NR-ECOL-TP-03-02. Available via http://www. terrestrial productivity: converging towards a single reo.gov/ecoshare/Publications/documents/FieldGuides/Co number? In: Roy J, Saugier B, Mooney HA (eds) Ter- ast/index.asp. Accessed 6 Feb 2008 restrial global productivity. Academic Press, San Diego, Merkle J (1951) An analysis of the plant communities of CA, pp 543–559 Mary’s Peak, Western Oregon. Ecology 32:618–640 Saunders DA, Hobbs RJ, Margules CR (1991) Biological Miller EA, Halpern CB (1998) Effects of environment and consequences of ecosystem fragmentation: a review. grazing disturbance on tree establishment in meadows of Conserv Biol 5:18–32 the central Cascade Range, Oregon, USA. J Veg Sci Schuller R, Exeter RL (2007) Grass Mountain research natural 9:265–282 area: guidebook supplement 32. General Technical Report Munger TT (1944) Out of the ashes of Nestucca. Am For PNW-GTR-732. Portland, OR. USDA Forest Service, 50:342–345, 366, 368 Pacific Northwest Research Station Oregon Division of Forestry (2001) Appendix H. History of the Snow BD (1984) Plant communities of the grassy balds of Northwest Oregon State Forests. In: Northwest Oregon state Marys Peak. Ph.D Dissertation, Oregon State University forests management plan. Available via http://www.odf. Thorson TD, Bryce SA, Lammers DA, Woods AJ, Omernik state.or.us/DIVISIONS/management/state_forests/sfplan/ JM, Kagan J, Pater DE, Comstock JA (2003) Ecoregions nwfmp01-final/nwfmp.asp. Accessed 20 Dec 2007 of Oregon (color poster with Geological Survey, map Oregon Natural Heritage Information Center [ONHIC] (2007) scale 1:1,500,000) Rare, threatened and endangered species of Oregon. Tillman D, May RM, Lehman CL, Nowak MA (2002) Habitat Oregon Natural Heritage Information Center. Available destruction and the extinction debt. Nature 371:65–66 via http://oregonstate.edu/ornhic/2007_t&e_book.pdf. Trombulak SC, Frissell CA (2000) Review of ecological Accessed 15 December 2007 effects of roads on terrestrial and aquatic communities. Oregon Plant Atlas version 2.0 (2007) Available via http://www. Conserv Biol 14:18–30 oregonflora.org/atlas.php. Accessed 23 May 2008 Turner MG, Gardner RH, O’Neill RV (2001) Landscape Parendes LA, Jones JA (2000) Role of light availability and ecology in theory and practice: pattern and process. dispersal in exotic plant invasion along roads and streams Springer, New York, NY in the H.J. Andrews experimental forest, Oregon. Conserv U.S. Fish and Wildlife Service [USFWS] (2001) Oregon sil- Biol 14:64–75 verspot butterfly (Speyeria zerene hippolyta) revised Reynolds SP (1993) Historical overview of the Alsea/Lobster recovery plan. U.S. Fish and Wildlife Servie, Portland, valley region of Benton County Oregon. Benton County Oregon. Available via http://www.fws.gov/pacific/eco Historical Society, Philomath, Oregon services/endangered/recovery/silverspot/. Accessed 22 Rochefort MR, Peterson DL (1996) Temporal and spatial dis- May 2008 tribution of trees in subalpine meadows of Mount Rainier Walker GW, MacLeod NS (1991) Geologic map of Oregon. National Park, Washington, U.S.A. Arct Alp Res 28:52– U.S. Department of the Interior, U.S. Geological Survey, 59 Reston, VA Rose R, Ketchum JS (2002) Interaction of vegetation control Wiegl PD, Knowles TW (1995) Megaherbivores and Southern and fertilization on conifer species across the Pacific Appalachian grass balds. Growth Change 26:365–382 Northwest. Can J For Res 32:136–152 Woodward A, Schreiner EG, Silsbee DG (1995) Climate, Rose R, Ketchum JS, Hanson DE (1999) Tree-year survival geography, and tree establishment in subalpine meadows and growth of Douglas-fir seedlings under various vege- of the Olympic Mountains, Washington, U.S.A. Arct Alp tation-free regimes. For Sci 45:117–126 Res 27:217–225

123